Land Use, Transit Flows and Demographic Patterns

Nutvara J, Kaitlyn Ng, Kittibhum Tasanasuwan

1. Background

Transportation policy-makers often face the challenge of choosing between various transportation scenarios. To do this, they need to understand the flow of origin and destination (OD) of people. King County has diverse land use, including urban areas like Seattle, small towns like Issaquah, and rural areas like Snoqualmie. With different land uses, the population demographics also vary. Therefore, understanding how demographics and land use affect travel patterns is crucial for urban planning and transportation infrastructure development.

This study aims to analyze the correlations between travel patterns, demographics, and land use. The study seeks to answer the following questions:

• What is the correlation between OD flows and demographics given similar land uses?
• How do land use and demographics affect the number of trips?
• Does the city provide a reasonable amount of transit to each census tract based on demographics and land use types?

2. Data

Our data came from three sources for King County in 2019.

1. Travel flow data from the Puget Sound Regional Council Household Travel Survey
   • Variables extracted: number of transit trips originating and ending in each census tract in King County
   • Workflow: created origin-destination matrix for census tracts in King County (sparsity issue), filtered data to only include transit trips
2. Land use data from General Land Use Final Dataset
   • Variables extracted: largest land use of each census tract
   • Workflow: clipped land use data by census tract geometries to determine the land use with the largest area proportion

5. Demographic data from *[American Community Survey](https://www.census.gov/programs-surveys/acs)* - Variables extracted: various demographic data (race, income, immigration status, housing costs, mode to travel to work, etc.) - Workflow: identified variables of interest, queried from Census API

Load and Merge Data

3. Analysis & Discussion

3.1 Correlation matrix between demographic and OD flow

Correlation coefficients are used to find how strong a relationship is between data. The formulas return a value between -1 and 1, where

• 1 indicates a strong positive relationship.
• -1 indicates a strong negative relationship.
• 0 indicates no relationship at all.

In this study, we want to observe the relationship between demographics and OD flows categorized by different land uses.

Data Preparations

Tract                             object
tractid                            int64
origin_flow                      float64
dest_flow                        float64
Name                              object
Density                          float64
% Race:WhiteAlone                float64
% Race:BlackAlone                float64
% Race:AsianAlone                float64
% Race:Other                     float64
% BelowPovertyLevel              float64
% Immigration                    float64
% Umemployment                   float64
% ToWork:Car                     float64
% ToWork:Transit                 float64
% ToWork:Bike                    float64
% ToWork:Walk                    float64
% ToWork:WFH                     float64
RatioMedianGrossRentToIncome     float64
RatioMedianHomeValueToIncome     float64
geometry                        geometry
land_use                          object
dtype: object

	origin_flow	dest_flow	Density	% Race:WhiteAlone	% Race:BlackAlone	% Race:AsianAlone	% Race:Other	% BelowPovertyLevel	% Immigration	% Umemployment	% ToWork:Car	% ToWork:Transit	% ToWork:Bike	% ToWork:Walk	% ToWork:WFH	RatioMedianGrossRentToIncome	RatioMedianHomeValueToIncome	land_use
0	272.95695	272.95695	1730.720957	52.318286	14.521806	14.322877	18.837031	9.854629	18.944147	2.876817	52.639633	5.554705	0.000000	0.336649	1.438409	0.022329	5.293762	Urban Character Residential
1	NaN	NaN	1239.464041	62.771662	8.947220	18.204911	10.076207	3.810330	4.826418	2.681343	48.461756	3.415185	0.000000	0.310471	3.358736	0.017208	3.646323	Natural Preservation and Conservation
2	NaN	NaN	1492.774647	59.607732	6.310404	12.734508	21.347356	7.561114	9.238204	1.705514	44.030699	2.046617	0.000000	0.255827	2.956225	0.020622	3.848660	Urban Character Residential
3	0.00000	0.00000	2078.075797	37.876106	22.973451	26.070796	13.079646	3.185841	11.469027	2.371681	45.451327	5.380531	0.000000	0.230088	3.238938	0.018121	4.529858	Undesignated
4	NaN	NaN	1968.115919	21.297284	25.056326	37.578560	16.067829	14.704139	18.024428	1.138385	41.088581	9.107079	0.166014	0.154156	1.921025	0.025035	6.767804	Intensive Urban

367

land_use
Urban Character Residential              218
Intensive Urban                          106
Industrial                                21
Rural Character Residential                9
Active Open Space and Recreation           7
Natural Preservation and Conservation      3
Public                                     3
Name: count, dtype: int64

land_use
Urban Character Residential    218
Intensive Urban                106
Other                           22
Industrial                      21
Name: count, dtype: int64

3.1.1 Urban Character Residential Area

In Urban Character Residential, OD flows are slightly higher correlated with the percentage of the population using transit and walking to work. This correlation suggests that more people choose transit and walking modes of transportation for their daily commute, leading to higher OD flows associated with them.

3.1.2 Intensive Urban Area

In Intensive Urban, OD flows are positively correlated with To Work by Walk, Ratio Rent to Income, and Ratio House Value to Income,and negatively correlated with Going To Work by Car. This indicates a preference for walking as a mode of transportation in these areas in contrast to driving. Also, if the cost of renting and house value relative to income is higher, there are increased OD flows.

3.1.3 Industrial Area

In Industrial Areas, All Transportation modes to work have strong positive correlations with OD flows, except car that is strong negative. Unemployment also shows a noticeable negative correlation with OD flows. If percentages of alternative modes are higher, the OD flows are higher. This reflects that population a reliance on alternative modes of transportation to work. However, if unemployment is lower, the OD flows are higher.

3.1.4 Other Areas

In Other Areas, the Percentage of Black race, the Percentage Below Poverty, and Rent/Income have strong positive correlations with OD flows. Driving to work* also shows a strong negative correlation with OD flows. This indicates that if areas are populated by Black people, below poverty, or transit users, their OD flows are higher as well.

3.2 OLS

Linear regression is a statistical technique used to describe relationships among variables. It can predict the relationship between variables by assuming a linear connection between the one or several independent variables (x) and dependent variable (y). The formula is given as:

Y = $B_0$ + $B_1$$X_1$ + $B_1$$X_2$ + ... + $B_p$$X_p$ + $\epsilon$

Where

• Y = the dependent or predicted variable
• $B_0$ = the y-intercept
• $B_1$ and $B_2$ = regression coefficients representing the change in y relative to a one-unit change in $X_1$ and $X_2$, respectively
• $B_p$ = the slope coefficient for each independent variable
• $\epsilon$ = the model’s random error (residual) term

This study uses regression analysis to answer the following questions

1. Does race relate with trip number?
2. What type of land use has impact to the number of trip origin and trip destination?
3. Which factor effect number of the trip more: unemployment or poverty level?
4. What type of land use has higher number of transit to work use? what about WFH?

3.2.0 Clean data

Remove rows with NaN value and land use data that is 'N/A' type or 'Undesignated' type

Inspecting land use type data distribution

3.2.1 Race VS Trip Number

The dependent variable is the sum of original flow and destination flow. The independent variables are percentage of each race in each census

/Users/kaitlynng/opt/anaconda3/lib/python3.9/site-packages/seaborn/axisgrid.py:2076: UserWarning: The `size` parameter has been renamed to `height`; please update your code.
  warnings.warn(msg, UserWarning)

OLS Regression Results

Dep. Variable:	y	R-squared:	0.023
Model:	OLS	Adj. R-squared:	0.008
Method:	Least Squares	F-statistic:	1.539
Date:	Mon, 04 Mar 2024	Prob (F-statistic):	0.206
Time:	12:06:31	Log-Likelihood:	-2118.7
No. Observations:	198	AIC:	4245.
Df Residuals:	194	BIC:	4258.
Df Model:	3
Covariance Type:	nonrobust

	coef	std err	t	P>|t|	[0.025	0.975]
% Race:WhiteAlone	49.7496	19.213	2.589	0.010	11.856	87.644
% Race:BlackAlone	141.2025	99.646	1.417	0.158	-55.326	337.731
% Race:AsianAlone	147.1504	56.369	2.610	0.010	35.976	258.325
% Race:Other	-80.6080	101.507	-0.794	0.428	-280.807	119.591

Omnibus:	198.933	Durbin-Watson:	2.009
Prob(Omnibus):	0.000	Jarque-Bera (JB):	4251.077
Skew:	3.975	Prob(JB):	0.00
Kurtosis:	24.262	Cond. No.	11.1

Notes:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.

A positive coefficient indicates that as the predictor variable increases, the Target variable also increases. From the result, as percentage of one race in census increase, the number of total trip also increase. Census with most majority is Asian trend to produce trip the most

Plots below show relationship of race dominant and number of trip origin

3.2.2 Trip Number vs Land Use Type

This analysis use original flow and destination flow as dependent value. The independent variables are land use type (categorical value)

	land_use_Industrial	land_use_Intensive Urban	land_use_Rural Character Residential	land_use_Urban Character Residential
0	0	0	0	1
9	0	1	0	0
12	0	0	0	0
13	1	0	0	0
14	0	0	0	1

OLS Regression Results

Dep. Variable:	y	R-squared (uncentered):	0.236
Model:	OLS	Adj. R-squared (uncentered):	0.220
Method:	Least Squares	F-statistic:	14.94
Date:	Mon, 04 Mar 2024	Prob (F-statistic):	1.16e-10
Time:	12:06:53	Log-Likelihood:	-2119.2
No. Observations:	198	AIC:	4246.
Df Residuals:	194	BIC:	4260.
Df Model:	4
Covariance Type:	nonrobust

	coef	std err	t	P>|t|	[0.025	0.975]
land_use_Industrial	5250.1097	2718.834	1.931	0.055	-112.158	1.06e+04
land_use_Intensive Urban	7460.4255	1208.371	6.174	0.000	5077.196	9843.655
land_use_Rural Character Residential	9779.1678	7690.023	1.272	0.205	-5387.615	2.49e+04
land_use_Urban Character Residential	4481.7169	1109.959	4.038	0.000	2292.580	6670.853

Omnibus:	191.464	Durbin-Watson:	2.040
Prob(Omnibus):	0.000	Jarque-Bera (JB):	3517.431
Skew:	3.816	Prob(JB):	0.00
Kurtosis:	22.186	Cond. No.	6.93

Notes:
[1] R² is computed without centering (uncentered) since the model does not contain a constant.
[2] Standard Errors assume that the covariance matrix of the errors is correctly specified.

Box plot shows that 'Intensive urban resident' and 'Urban Characteristic Resident' has wide range of the number of trip, however, most of data are outliner which indicated that there are some special area where there are trip generated more than usual. From regression analysis, it can indicate that rural residential area are likely to generate amouth of trips. This make sense as facilities might be limited in rural area so people need to make trip to do their activities

3.2.3 Which factor effect number of the trip more: unemployment or poverty level?

This analysis use original flow and destination flow as dependent value. The independent variables are percentage of unemployment population and percentage of population with BelowPovertyLevel

Text(0.5, 1.0, '% Unemployment vs % BelowPovertyLevel')

OLS Regression Results

Dep. Variable:	y	R-squared (uncentered):	0.234
Model:	OLS	Adj. R-squared (uncentered):	0.226
Method:	Least Squares	F-statistic:	29.89
Date:	Mon, 04 Mar 2024	Prob (F-statistic):	4.69e-12
Time:	12:07:05	Log-Likelihood:	-2119.4
No. Observations:	198	AIC:	4243.
Df Residuals:	196	BIC:	4249.
Df Model:	2
Covariance Type:	nonrobust

	coef	std err	t	P>|t|	[0.025	0.975]
% Umemployment	894.5723	461.404	1.939	0.054	-15.381	1804.526
% BelowPovertyLevel	282.8231	108.976	2.595	0.010	67.907	497.739

Omnibus:	190.443	Durbin-Watson:	2.015
Prob(Omnibus):	0.000	Jarque-Bera (JB):	3486.356
Skew:	3.785	Prob(JB):	0.00
Kurtosis:	22.112	Cond. No.	7.84

Notes:
[1] R² is computed without centering (uncentered) since the model does not contain a constant.
[2] Standard Errors assume that the covariance matrix of the errors is correctly specified.

From scatter plot, percentage of unemployment and population with below poverty level are related. However, these two factor has different effect to the number of trip. From linear regression analysis, area with higher unemployment makes more trip than area with people below poverty level. This can be concluded that people below poverty cannot affroad travel expense than unemployememt people

3.2.4 What type of land use has higher number of transit to work use? what about WFH?

This analysis use percentage of people who use transit to work or percentage of people who work from home (WFH) as dependent value. The independent variables are land use type (categorial)

OLS Regression Results

Dep. Variable:	% ToWork:Transit	R-squared (uncentered):	0.792
Model:	OLS	Adj. R-squared (uncentered):	0.788
Method:	Least Squares	F-statistic:	184.6
Date:	Mon, 04 Mar 2024	Prob (F-statistic):	5.77e-65
Time:	12:07:23	Log-Likelihood:	-611.87
No. Observations:	198	AIC:	1232.
Df Residuals:	194	BIC:	1245.
Df Model:	4
Covariance Type:	nonrobust

	coef	std err	t	P>|t|	[0.025	0.975]
land_use_Industrial	12.0806	1.343	8.992	0.000	9.431	14.730
land_use_Intensive Urban	12.2899	0.597	20.583	0.000	11.112	13.468
land_use_Rural Character Residential	7.0098	3.800	1.845	0.067	-0.484	14.504
land_use_Urban Character Residential	8.3243	0.548	15.178	0.000	7.243	9.406

Omnibus:	4.138	Durbin-Watson:	1.520
Prob(Omnibus):	0.126	Jarque-Bera (JB):	4.141
Skew:	0.320	Prob(JB):	0.126
Kurtosis:	2.694	Cond. No.	6.93

Notes:
[1] R² is computed without centering (uncentered) since the model does not contain a constant.
[2] Standard Errors assume that the covariance matrix of the errors is correctly specified.

OLS Regression Results

Dep. Variable:	% ToWork:WFH	R-squared (uncentered):	0.799
Model:	OLS	Adj. R-squared (uncentered):	0.795
Method:	Least Squares	F-statistic:	192.8
Date:	Mon, 04 Mar 2024	Prob (F-statistic):	2.04e-66
Time:	12:07:27	Log-Likelihood:	-411.41
No. Observations:	198	AIC:	830.8
Df Residuals:	194	BIC:	844.0
Df Model:	4
Covariance Type:	nonrobust

	coef	std err	t	P>|t|	[0.025	0.975]
land_use_Industrial	2.8205	0.488	5.778	0.000	1.858	3.783
land_use_Intensive Urban	4.0694	0.217	18.758	0.000	3.642	4.497
land_use_Rural Character Residential	2.2782	1.381	1.650	0.101	-0.445	5.001
land_use_Urban Character Residential	3.9001	0.199	19.572	0.000	3.507	4.293

Omnibus:	7.055	Durbin-Watson:	1.795
Prob(Omnibus):	0.029	Jarque-Bera (JB):	6.978
Skew:	0.457	Prob(JB):	0.0305
Kurtosis:	3.096	Cond. No.	6.93

Notes:
[1] R² is computed without centering (uncentered) since the model does not contain a constant.
[2] Standard Errors assume that the covariance matrix of the errors is correctly specified.

Percentage of people who WFH is lower than people who transit to work in every land use type. People mostly use transit to work in Industrial and Intencive Urban area. As these area mostly have good transit facility and mainly serving for people who working. However, people who live in intenseive urban area tend to WFH the most. This indicated that intensive urban area must have both office and residential area.

3.3 Mapping Analysis

These graphs will provide insight on the relationship between OD flows, demographics, and land use.

What is the relationship between transit flow patterns and demographics?

Transit OD flows are plotted via scikit-mobility, which is a Python library for human mobility analysis. Their visualizations are built on top of folium.

Make this Notebook Trusted to load map: File -> Trust Notebook

Transit OD patterns:

• We observe large flows between Kirkland and Woodinville (locations with large transit stations)
• Many flows come in and out of downtown, even from as far as Federal Way and from the water taxi to Vashon Island
• Important nodes in network appear to be downtown, UW/U District, Bellevue, and the Microsoft campus in Redmond

OD relationship to demographics:

• There appears to be a slight correlation with % Black, lower-income tracts and larger OD flows. From the map, this is not clear if this is due to Black and/or lower-income populations being more reliant on transit, or whether transit stations are being built in areas with larger Black/lower-income areas due to cheaper construction costs. It should also be noted that transit-oriented development tends to gentrify neighborhoods, so these correlations may change over time as neighborhoods gentrify.

For the top three land uses, are there differences between demographics and transit flows?

Land Use Type	Number of Tracts
Urban Character Residential	218
Intensive Urban	106
Industrial	21

Warning: Cannot change to a different GUI toolkit: notebook. Using widget instead.

Observations:

• A disproportionate amount of Black, low-income, and unemployed individuals live in areas with industrial land uses, or in census tracts adjacent to industrial land uses. This is a reflection of historical redlining practices in Seattle which disadvantaged people of color, and the implications are still present today. There are also signficiant environmental justice concerns (e.g., air quality) living in industrial areas.
• Transit serves intensive urban areas (e.g., Seattle downtown) well, which is reflected by high ridership in those areas. Unsurprisingly, the further from downtown core, the fewer commuters via transit.
• Interestingly enough, very few individuals who live in Industrial areas are unable to work from home. These individuals may be in service jobs where they do not have the flexibility to choose their work location (e.g., manufacturing jobs, service jobs).

4. Conclusion

[NbConvertApp] Converting notebook presentation_notebook.ipynb to html
[NbConvertApp] WARNING | Alternative text is missing on 13 image(s).
[NbConvertApp] Writing 27707026 bytes to presentation_notebook.html

From our analysis, several conclusions can be drawn as follows:

1) OD flows and demographics

• A slight correlation exists between % Black, lower-income tracts, and larger OD flows, suggesting potential reliance on transit or station construction in areas with lower costs.

2) Land use and number of trips

• In Industrial and Intensive Urban areas, people mostly use transit for commuting, especially those with service jobs that lack the flexibility to work from home.

3) Justify the amount of transit provided

• In Intensive Urban areas, people prefer walking over driving, suggesting the presence of adequate facilities for walking.
• Significant flows occur in and out of downtown areas, indicating high ridership and effective transit service in these intensive urban locations